Code for paper "Continuous-time identification of dynamic state-space models by deep subspace encoding"

See environment.yml to which packages are required. (see Managing envirments with conda)

The only package this is not installed automatically when using environment.yml is deepSI (version 3.13). This package should be installed manually using the instruction on that github page.

The important notebooks are;

• encoder-CT-train.ipynb used to train the CT subnet
• encoder-CT-analysis.ipynb used to analyze the results obtained from encoder-CT-train.ipynb
• latent-neural-ode-train.ipynb used to train the ODE with exogenous inputs.
• latent-neural-ode-analysis.ipynb used to analyze the results obtained from latent-neural-ode-train.ipynb
• EMPS-train-and-analysis.ipynb used for training and analysis for both neural ODE and CT subnet on the EMPS benchmark

The considered benchmarks are described and downloaded from: https://www.nonlinearbenchmark.org/

To run the the analysis you need either the estimate model or re-train using the train notebooks. The estimated model can be downloaded from: drive (426 MB)